A simple scoring system based on clinical features to predict locally advanced rectal cancers

Abstract

Purpose: The purpose of this study was to identify clinical risk factors and establish a prediction scoring system for locally advanced rectal cancer.

Materials and methods: Retrospective univariate and multivariate logistic analyses were conducted for 413 curable rectal cancer patients. Clinical factors found to be significantly related with tumor stages were incorporated into a scoring system to predict locally advanced stages, which was validated in an independent cohort of 279 rectal cancer patients.

Results: In the training set, tumor size, differentiation, and serum carcinoembryonic antigen (CEA) level (P < 0.01) were significant predictors of locally advanced rectal cancer in both univariate and multivariate analyses, which were incorporated into a proposed scoring system to predict locally advanced stages. The area under the receiver operating characteristic curve (AUROC) of this scoring system was 0.751 and the prediction accuracy was 78.2%. Patients were categorized into three subsets according to the total score. The low-risk group (score 0) had a smaller chance (18.2%) to have locally advanced rectal cancer, compared to mean 49.2% for the intermediate-risk group (score 1) and mean 83.0% for the high-risk group (score of 2-4; P < 0.05). In the validation set, the AUROC of the scoring system was 0.756 and the prediction accuracy was 75.3%.

Conclusions: Tumor size more than 2 cm, poor differentiation, and elevated serum CEA level are high-risk factors of locally advanced rectal cancer. A simple scoring system based on these three factors may be valuable to predict locally advanced rectal cancer.